Review on FPGA-Based Accelerators in Deep learning
Yuhao Wu
Abstract
Convolutional neural network (CNN) has achieved outstanding performance, but a substantial computational burden limits its application. This paper introduces the common methods of FPGA-based accelerators in deep learning, especially the methods for CNN. The research status of FPGA accelerators is summarized from three aspects: hardware structure, fast convolution, and optimization strategy. Finally, challenges of accelerating deep learning on FPGAs are analyzed.
Topics & Concepts
Field-programmable gate arrayDeep learningComputer scienceConvolutional neural networkConvolution (computer science)Computer architectureArtificial intelligenceArtificial neural networkEmbedded systemComputer engineeringMachine learningParallel computingImage Processing Techniques and ApplicationsCCD and CMOS Imaging SensorsAdvanced Neural Network Applications